Aug 2026· PLoS ONE· Vol 21· 0 citations· 44 references
Medicine
TL;DR
Experimental results demonstrate that the integrated system improves locomotion stability, energy efficiency, and terrain adaptability compared with baseline controllers, highlighting the effectiveness of combining a structured gait prior, lightweight residual coordination, and hardware-aware deployment for practical quadruped locomotion.
Abstract
Quadruped robots have gained significant attention due to their superior mobility on uneven and unstructured terrains, offering potential applications in inspection, search and rescue, and field exploration. However, achieving robust locomotion on low-cost platforms remains challenging because of constraints in stability, adaptability, sensing quality, and onboard computation. In this work, we present an integrated motion-control framework that combines biologically inspired Central Pattern Generators (CPGs), a multi-agent reinforcement learning coordination layer, and low-cost hardware adaptation to enable reliable and efficient quadruped locomotion. The proposed framework uses CPGs as structured gait priors for rhythmic leg motion, models each leg as a coordinated agent with a shared-parameter residual policy, and incorporates actuator abstraction and safety-aware command projection. The low-cost merit specifically concerns online deployment: the four legs share a single 39,560-parameter actor (approximately 155 KiB in 32-bit precision), evaluated at 50 Hz from compact proprioceptive observations, while the centralized critic, simulation infrastructure, external motion capture, vision-based terrain perception, direct torque sensing, and online dynamics optimization are not required on the robot. We validate the approach in both simulation and on a physical low-cost quadruped robot across obstacles, ramps, stairs, and uneven terrain. Experimental results demonstrate that the integrated system improves locomotion stability, energy efficiency, and terrain adaptability compared with baseline controllers, highlighting the effectiveness of combining a structured gait prior, lightweight residual coordination, and hardware-aware deployment for practical quadruped locomotion.
Quadruped robots possess outstanding terrain adaptability and boast extensive application prospects in scenarios such as search and rescue, field exploration, and more. Nevertheless, conventional model-based motion control methods suffer from cumbersome modeling processes and poor generalization performance, making them ill-suited for unstructured complex environments. To address these limitations, this paper presents a comprehensive review of deep reinforcement learning-based motion control technologies for quadruped robots. It first organizes the fundamental theories concerning robot kinematics and reinforcement learning, then categorizes and summarizes research advances across three core research branches: gait generation, autonomous navigation, and adaptive gait transition. Furthermore, this paper analyzes prevailing challenges and corresponding countermeasures regarding hardware deployment, sample efficiency, and model generalization capacity. It points out that further integration of multi-algorithms, optimization of sim-to-real transformation and overall strategy design will be the main trends in this field. By identifying current technical bottlenecks and forecasting future development trends, this work offers valuable references for practical technical implementation and subsequent research within this field.
Hardware failures require legged robots to rapidly reorganize coordination and gait timing to maintain stability and mobility. This is particularly challenging for larger quadrupeds, where increased mass and tighter actuation limits reduce the feasibility of aggressive, high-frequency compensation strategies often observed on smaller platforms. In this work, we propose a deep reinforcement learning approach for fault-tolerant locomotion under actuator power loss. The method employs an asymmetric actor-critic architecture in which the critic has access to privileged information during training, while the actor learns to reconstruct a corresponding latent representation from proprioceptive observations. We introduce a latent-alignment loss that encourages consistency between actor and critic representations. Additionally, we augment the action space with a learnable gait frequency parameter, enabling adaptive gait timing in response to terrain variations and actuator degradation without predefined faulty-leg strategies. The approach is validated in high-fidelity simulation on uneven terrain and real-world experiments on flat ground using a 68 kg quadruped robot.
Giovanbattista Gravina, Luca Rossini, Carlo Rizzardo et al.· 0 citations
Deploying learned control policies on low-cost robotic platforms introduces transport latencies and noisy motor feedback that systematically widens the sim-to-real gap. The chasm of simulation to deployment in hardware lies in the delay of the actuator reaching the commanded position. On platforms such as the Mini Pupper 2, a measured>50 ms transport delay transforms the locomotion task from a standard Markov decision process into a partially observable one. In this paper, we take a biologically inspired approach of handling noisy and delayed feedback to close the sim-to-real gap, thereby expanding the capability of reinforcement learning on cost-constrained hardware. Using a low-cost quadrupedal hardware platform, we find that using a forward model of the average actuator delay, paired with a time-aware neural network results in robust locomotion. Additionally, our time-aware neural network learned a central pattern generator (CPG): a self-sustaining rhythmic gait that is robust to +320 ms latency perturbations, mirroring the CPGs found in the spinal cords of vertebrates. We posit that temporal self-organization may be a general strategy for cost-constrained locomotion.
Javier C. Weddington, Bence P. Ölveczky, S. Baccus· arXiv.org· 0 citations
Deep reinforcement learning has enabled quadrupedal robots to traverse challenging terrains, yet energy efficiency remains a limiting factor for prolonged autonomous operation. Most existing frameworks rely on fixed or adaptively tuned proportional-derivative (PD) controllers that operate exclusively in the joint space. Such approaches typically lack an explicit mechanism for contact compliance, often applying excessive torque on benign terrains while providing insufficient absorption of reaction forces on irregular surfaces. To address these limitations, we propose HIP, a hybrid impedance and PD control framework that fuses joint-space PD control for trajectory tracking with task-space impedance control for contact compliance. The impedance term, mapped to joint torques via the Jacobian transpose, models compliant foot-tip behavior that absorbs impact energy during ground contact rather than resisting it through rigid control. To coordinate the two control modalities, we further introduce the attention for representation combiner (ARC) network. The ARC network employs a cross-attention mechanism between a gain actor and a joint actor, enabling control gains and desired joint positions to be generated in a coordinated manner. A state estimator augmented with a per-leg stumble estimator provides additional proprioceptive context to both actors for proactive gain adaptation. Simulation experiments across diverse terrains demonstrate that HIP achieves velocity tracking accuracy comparable to existing baselines while delivering improved energy efficiency. Torque decomposition analysis further confirms that the impedance component effectively reduces torque peaks during contact events.
Hyeonwoo Lee, Mincheol Kim, Hyun Myung· 2026 23rd International Conf...· 0 citations
Quadruped robots have attracted increasing attention because they can traverse uneven terrain, support field deployment, and perform tasks that are difficult for wheeled or tracked platforms. Recent advances in artificial intelligence (AI) have further expanded their capabilities from manually designed gait control toward learning-based locomotion, perception-aware adaptation, dynamic motion skills, autonomous recovery, manipulation, energy-aware operation, fault diagnosis, and human–robot interaction. However, the literature on AI-driven quadruped robotics is distributed across diverse technical topics, robot platforms, validation settings, and performance metrics, making it difficult to assess the maturity and practical value of different approaches. To address this need, this review provides an AI-centered and deployment-oriented overview of quadruped robotics. A systematic literature search was conducted using Web of Science, IEEE Xplore, ACM Digital Library, ScienceDirect, and SpringerLink, covering studies published approximately from 2000 to 2025. After screening and eligibility assessment, 287 studies were included for detailed review. The review first examines AI-driven locomotion, including reinforcement learning, non-RL machine-learning methods, model-based approaches, and hybrid strategies, with attention to robustness, sim-to-real transfer, sensor use, computational requirements, and hardware validation. It then summarizes AI-supported advanced behaviors, including jumping, fall prevention and recovery, and object manipulation, focusing on reported quantitative performance, impact management, and reliability. Finally, it discusses system-level topics that affect real-world deployment, including fault diagnosis, energy-efficient control, shared autonomy, trust-aware and explainable interaction, and safety-aware human–robot collaboration. By organizing the literature according to robot capabilities, validation maturity, and deployment challenges, this review helps clarify the current progress, limitations, and future directions of AI-driven quadruped robots.
Li-Kai Wu, C. Chow, W. Wong et al.· Frontiers in Neurorobotics· 0 citations
Bipedal robots have gained a lot of attention in robotics because of their versatility in numerous application areas such as in rescue missions, military tasks, therapy, personal assistance and care for the aged. Their ability to move in tough and mixed-up spaces provides essential advantages over wheeled robots. However, they present challenges, such as complicated movement, balancing and control issues, and energy management problems. This paper utilized PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) to examine peer-reviewed publications in bipedal robots, covering both their modeling dynamics and strategic control, application areas and the challenges in development and management. After reviewing 165 of 15,214 publications that satisfied the inclusion criteria, the assessments were developed around the history of bipedal robots, marking important developments like dynamic control, passive-dynamic walking, and using sensors in real-time. Developments in artificial intelligence, machine learning, and reinforcement learning have made bipedal robots more stable, adaptable, and efficient. The novelty of this review lies in integrating rigid-body and reduced-order dynamic models, classical and learning-based control strategies, hardware limitations, and future research priorities within a unified comparative framework for bipedal robotics.
B. Kommey, E. Tamakloe, Safianu Umar et al.· AVITEC· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
